Revisiting the Tic Tac Toe Game

The easiest way to understand HyperState is by example. If you you did not see the Tic-Tac-Toe example, then please review it now, as we are going to use this to demonstrate how to use the Hyperstack::State::Observable module.

In our original Tic-Tac-Toe implementation the state of the game was stored in the DisplayGame component. State was updated by "bubbling up" events from lower level components up to DisplayGame where the event handler updated the state.

This is a nice simple approach but suffers from two issues:

• Each level of lower level components must be responsible for bubbling up the events to the higher component.

• The DisplayGame component is responsible for both managing state and displaying the game.

As our applications become larger we will want a way to keep each component's interface isolated and not dependent on the overall architecture, and to insure good separation of concerns.

The Hyperstack::State::Observable module allows us to put the game's state into a separate class which can be accessed from any component: No more need to bubble up events, and no more cluttering up our DisplayGame component with state management and details of the game's data structure.

Here is the game state and associated methods moved out of the DisplayGame component into its own class:

class Game
  include Hyperstack::State::Observable
​
  def initialize
    @history = [[]]
    @step = 0
  end
​
  observer :player do
    @step.even? ? :X : :O
  end
​
  observer :current do
    @history[@step]
  end
​
  state_reader :history
​
  WINNING_COMBOS = [[0, 1, 2], [3, 4, 5], [6, 7, 8], [0, 3, 6], [1, 4, 7], [2, 5, 8], [0, 4, 8], [2, 4, 6]]
​
  def current_winner?
    WINNING_COMBOS.each do |a, b, c|
      return current[a] if current[a] && current[a] == current[b] && current[a] == current[c]
    end
    false
  end
​
  mutator :handle_click! do |id|
    board = history[@step]
    return if current_winner? || board[id]
​
    board = board.dup
    board[id] = player
    @history = history[0..@step] + [board]
    @step += 1
  end
​
  mutator(:jump_to!) { |step| @step = step }
end

Let's go over the each of the differences from the code that was in the DisplayGame component.

class Game
  include Hyperstack::State::Observable

Game is now in its own class and includes Hyperstack::State::Observable. This adds a number of methods to Game that allows our class to become a reactive store. When Game interacts with other stores and components they will be updated as the state of Game changes.

  def initialize
    @history = [[]]
    @step = 0
  end

In the original implementation we initialized the two state variables @history and @step in the before_mount callback. The same initialization is now in the initialize method which will be called when a new instance of the game is created. This will still be done in the DisplayGame before_mount callback (see below.)

  observer :player do
    @step.even? ? :X : :O
  end
​
  observer :current do
    @history[@step]
  end

In the original implementation we had instance methods player and current. Now that Game is a separate class we define these methods using observer.

The observer method creates a method that is the inverse of mutator. While mutate (and mutator) indicate that state has been changed observe and observer indicate that state has been accessed outside the class.

  attr_reader :history

Just as we have mutate, mutator, and state_writer, we have observe, observer, and state_reader.

  WINNING_COMBOS = [[0, 1, 2], [3, 4, 5], [6, 7, 8], [0, 3, 6], [1, 4, 7], [2, 5, 8], [0, 4, 8], [2, 4, 6]]
​
  def current_winner?
    WINNING_COMBOS.each do |a, b, c|
      return current[a] if current[a] && current[a] == current[b] && current[a] == current[c]
    end
    false
  end

We don't need any changes to current_winner?. It accesses the internal state through the current method so there is no need to explicitly make current_winner? an observer (but we could, without affecting anything.)

  mutator :handle_click! do |id|
    board = history[@step]
    return if current_winner? || board[id]
​
    board = board.dup
    board[id] = player
    @history = history[0..@step] + [board]
    @step += 1
  end
​
  mutator(:jump_to!) { |step| @step = step }
end

Finally we need no changes to the handle_click! and jump_to! mutators either.

The Updated DisplayGame Component

class DisplayGame < HyperComponent
  before_mount { @game = Game.new }
  def moves
    return unless @game.history.length > 1
​
    @game.history.length.times do |move|
      LI(key: move) { move.zero? ? "Go to game start" : "Go to move ##{move}" }
        .on(:click) { @game.jump_to!(move) }
    end
  end
​
  def status
    if (winner = @game.current_winner?)
      "Winner: #{winner}"
    else
      "Next player: #{@game.player}"
    end
  end
​
  render(DIV, class: :game) do
    DIV(class: :game_board) do
      DisplayCurrentBoard(game: @game)
    end
    DIV(class: :game_info) do
      DIV { status }
      OL { moves }
    end
  end
end

The DisplayGame before_mount callback is still responsible for initializing the game, but it no longer needs to be aware of the internals of the game's state. It simply calls Game.new and stores the result in the @game instance variable. For the rest of the component's code we call the appropriate method on @game.

We will need to pass the entire game to DisplayBoard (we will see why shortly) so we will rename it to DisplayCurrentBoard.

As we will see DisplayCurrentBoard will be responsible for directly notifying the game that a user has clicked, so we do not need to handle any events coming back from DisplayCurrentBoard.

The DisplayCurrentBoard Component

class DisplayCurrentBoard < HyperComponent
  param :game
​
  def draw_square(id)
    BUTTON(class: :square, id: id) { game.current[id] }
    .on(:click) { game.handle_click!(id) }
  end
​
  render(DIV) do
    (0..6).step(3) do |row|
      DIV(class: :board_row) do
        (row..row + 2).each { |id| draw_square(id) }
      end
    end
  end
end

The DisplayCurrentBoard component receives the entire game, and it will access the current board, using the current method, and will directly notify the game when a user clicks using the handle_click! method.

By having DisplayCurrentBoard directly deal with user actions, we simplify both components as they do not have to communicate back upwards via events. Instead we communicate through the central game store.

The Flux Loop

Rather than sending params down to lower level components, and having the components bubble up events, we have created a Flux Loop. The Game store holds the state, the top level component reads the state and sends it down to lower level components, those components update the Game state causing the top level component to re-rerender.

This structure greatly simplifies the structure and understanding of our components, and keeps each component functionally isolated.

Furthermore algorithms such as current_winner? now are neatly abstracted out into their own class.

Classes and Instances

If we are sure we will only want one game board, we could define Game with class methods like this:

class Game
  include Hyperstack::State::Observable
​
  class << self
    def initialize
      @history = [[]]
      @step = 0
    end
​
    observer :player do
      @step.even? ? :X : :O
    end
​
    observer :current do
      @history[@step]
    end
​
    state_reader :history
​
    WINNING_COMBOS = [[0, 1, 2], [3, 4, 5], [6, 7, 8], [0, 3, 6], [1, 4, 7], [2, 5, 8], [0, 4, 8], [2, 4, 6]]
​
    def current_winner?
      WINNING_COMBOS.each do |a, b, c|
        return current[a] if current[a] && current[a] == current[b] && current[a] == current[c]
      end
      false
    end
​
    mutator :handle_click! do |id|
      board = history[@step]
      return if current_winner? || board[id]
​
      board = board.dup
      board[id] = player
      @history = history[0..@step] + [board]
      @step += 1
    end
​
    mutator(:jump_to!) { |step| @step = step }
  end
end
​
class DisplayBoard < HyperComponent
  param :board
​
  def draw_square(id)
    BUTTON(class: :square, id: id) { board[id] }
    .on(:click) { Game.handle_click!(id) }
  end
​
  render(DIV) do
    (0..6).step(3) do |row|
      DIV(class: :board_row) do
        (row..row + 2).each { |id| draw_square(id) }
      end
    end
  end
end
​
class DisplayGame < HyperComponent
  def moves
    return unless Game.history.length > 1
​
    Game.history.length.times do |move|
      LI(key: move) { move.zero? ? "Go to game start" : "Go to move ##{move}" }
        .on(:click) { Game.jump_to!(move) }
    end
  end
​
  def status
    if (winner = Game.current_winner?)
      "Winner: #{winner}"
    else
      "Next player: #{Game.player}"
    end
  end
​
  render(DIV, class: :game) do
    DIV(class: :game_board) do
      DisplayBoard(board: Game.current)
    end
    DIV(class: :game_info) do
      DIV { status }
      OL { moves }
    end
  end
end

Now instead of creating an instance and passing it around we call the class level methods on Game throughout.

The Hyperstack::State::Observable module will call any class level initialize methods in the class or subclasses before the first component mounts.

Note that with this approach we can go back to passing just the current board to DisplayBoard as DisplayBoard can directly access Game.handle_click! since there is only one game.

Thinking About Stores

To summarize: a store is simply a Ruby object or class that using the observe and mutate methods marks when its internal data has been observed by some other class, or when its internal data has changed.

When components render they observe stores throughout the system, and when those stores mutate the components will rerender.

You as the programmer need only to remember that public methods that read internal state must at some point during their execution declare this using observe, observer, state_reader or state_accessor methods. Likewise a method that changes internal state must declare this using mutate, mutator, state_writer or state_accessor methods.

If your store's methods access other stores, you do not need worry about their state, only your own. On the other hand keep in mind that the built in Ruby Array and Hash classes are not stores, so when you modify or read an Array or a Hash its up to you to use the appropriate mutate or observe method.

Stores and Parameters

Typically in a large system you will have one or more central stores, and what you end up passing as parameters are either instances of those stores, or some other kind of index into the store. If there is only one store (as in the case of our Game), you need not pass any parameters at all.

We can rewrite the previous iteration of DisplayBoard to demonstrate this:

class DisplaySquare
  param :id
  render
    BUTTON(class: :square, id: id) { Game.current[id] }
    .on(:click) { Game.handle_click(id) }
  end
end
​
class DisplayBoard < HyperComponent
  render(DIV) do
    (0..6).step(3) do |row|
      DIV(class: :board_row) do
        (row..row + 2).each { |id| DisplaySquare(id: id) }
      end
    end
  end
end

Here DisplayBoard no longer takes any parameter (and could be renamed again to DisplayCurrentBoard) and now a new component - DisplaySquare - takes the id of the square to display, but the game or the current board are never passed as parameters; there is no need to as they are implicit.

Whether to pass (or not pass) a store class, an instance of a store, or some other index into the store is a design decision that depends on lots of factors, mainly how you see your application evolving over time.

Summary of Methods

All the observable methods can be used either at the class or instance level.

Observing State: observe, observer, state_reader

The observe method takes any number of arguments and/or a block. The last argument evaluated or the value of the block is returned.

The arguments and block are evaluated then the object's state will be observed.

If the block exits with a return or break, the state will not be observed.

# evaluate and return a value
observe @history[@step]
​
# evaluate a block and return its value
observe do
  @history[@step]
end

The observer method defines a new method with an implicit observe:

observer :foo do |x, y, z|
  ...
end

is equivilent to

def foo(x, y, z)
  observe do
    ...
  end
end

Again if the block exits with a return or break the state will not be observed.

The state_reader method declares one or more state accessors with an implicit state observation:

state_reader :bar, :baz

is equivilent to

def bar
  observe @bar
end
def baz
  observe @baz
end

Mutating State: mutate, mutator, state_writer, toggle

The mutate method takes any number of arguments and/or a block. The last argument evaluated or the value of the block is returned.

The arguments and block are evaluated then the object's state will be mutated.

If the block exits with a return or break, the state will not be mutated.

# evaluate and return a value
mutate @history[@step]
​
# evaluate a block and return its value
mutate do
  @history[@step]
end

The mutator method defines a new method with an implicit mutate:

mutator :foo do |x, y, z|
  ...
end

is equivilent to

def foo(x, y, z)
  mutate do
    ...
  end
end

Again if the block exits with a return or break the state will not be mutated.

The state_writer method declares one or more state accessors with an implicit state mutation:

state_reader :bar, :baz

is equivilent to

def bar=(x)
  mutate @bar = x
end
def baz=(x)
  observe @baz = x
end

The toggle method reverses the polarity of a instance variable:

toggle(:foo)

is equivilent to

mutate @foo = !@foo

The state_accessor Method

Combines state_reader and state_writer methods.

state_accessor :foo, :bar

is equivilent to

state_reader :foo, :bar
state_writer :foo, :bar

Components and Stores

The standard HyperComponent base class includes Hyperstack::State::Observable so any HyperComponent has access to all of the above methods. A component also always observes itself so you never need to use observe within a component unless the state will be accessed outside the component. However once you start doing that you would be better off to move the state into a separate store.

In addition components also act as the Observers in the system. What this means is that the current component that is running its render method is recording all stores that call observe, when a store mutates, then all the components that recorded observations will be rerendered.